lable at ScienceDirect

Water Research 121 (2017) 302e310

Contents lists avai

Water Research

journal homepage: www.elsevier .com/locate/watres

Optimization of sulfate removal from brackish water by membranecapacitive deionization (MCDI)

Wangwang Tang, Di He**, Changyong Zhang, T. David Waite*

School of Civil and Environmental Engineering, University of New South Wales, Sydney, NSW 2052, Australia

a r t i c l e i n f o

Article history:Received 19 March 2017Received in revised form15 May 2017Accepted 22 May 2017Available online 22 May 2017

Keywords:Membrane capacitive deionizationSulfate removalProcess optimizationPreferential adsorptionWater recoveryEnergy consumption

* Corresponding author.** Corresponding author.

E-mail addresses: di.he@unsw.edu.au (D. He), d.wa

http://dx.doi.org/10.1016/j.watres.2017.05.0460043-1354/© 2017 Elsevier Ltd. All rights reserved.

a b s t r a c t

Removal of sulfate from water is an environmental challenge faced by many industrial sectors as mostexisting options are inefficient, costly or unsustainable. The situation is further complicated by the typicalcoexistence of other ions. In this work, the feasibility of sulfate removal from brackish water by single-pass constant-current membrane capacitive deionization (MCDI) under reverse-current desorption wasinvestigated. Results revealed that sulfate is preferentially removed from the aqueous solution by MCDIcompared to chloride. Equivalent circuits of the MCDI system during adsorption and desorption wereproposed and the dynamic variation of cell voltage and charging voltage at different adsorption currentswas satisfactorily elucidated. Optimization studies were conducted with attention given to discussing theeffects of four operating parameters, i.e., adsorption current, pump flow rate, ending cell voltage anddesorption current, on three performance indicators (i.e., water recovery, energy consumption andsorption ratio of sulfate to chloride) on the premise of maintaining the effluent sulfate concentrationbelow the specified threshold of 300 mg L�1. Water recovery-energy consumption mapping and sorptionratio of sulfate to chloride-energy consumption mapping indicated that the combination of a loweradsorption current and a lower matching pump flow rate which reduced the effluent sulfate concen-tration to 300 mg L�1 was more favorable in practical applications. An appropriately small ending cellvoltage was advantageous while a trade-off between water recovery and energy cost was required inoptimizing the desorption current.

© 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Sulfate (SO42�) is ubiquitous not only in natural waters, but also

in municipal and industrial wastewaters (C�alinescu et al., 2016;Runtti et al., 2016). Although sulfate is not usually considered ahealth concern, high concentrations of sulfate in drinking water cancause a bitter taste and, when its concentration exceeds 600mg L�1,diarrhea (Guimar~aes and Le~ao, 2014). Discharge of high levels ofsulfate can significantly affect public water supplies by causingcorrosion and/or scaling of pipes and equipment. Moreover,hydrogen sulfide (H2S), which is toxic to the ecosystem, could beproduced through sulfate reduction by sulfate-reducing bacteriaunder anaerobic conditions (Chen et al., 2016). Because of theseadverse effects to human health and the environment, manycountries have set the maximum sulfate concentration values

ite@unsw.edu.au (T.D. Waite).

ranging from 250 mg L�1 to 500 mg L�1 depending on the end useof the water source (Runtti et al., 2016). In addition, it is importantto remove sulfate from recycled wastewater when transferring theconcentrate from the membrane separation process back to theinitial biological system (such as a membrane bioreactor) in orderto minimize the effects of sulfate on both the efficacy of the bio-logical process and membrane scale formation. To date, severalmethods have been applied in treating waters containing excessivesulfate, including chemical precipitation, adsorption and/or ionexchange, biological treatment and reverse osmosis. However,these methods suffer from limitations. For example, precipitationand adsorption methods produce large amounts of potentially toxicsludges. Biological methods require significant amounts of organiccarbon (as bacterial feed) and careful anaerobic sludge manage-ment (Silva et al., 2012). Reverse osmosis methods are costlydespite enabling treatment of large volumes of waters. As such,seeking a more cost-effective, sustainable and easy-to-operatealternative method for sulfate removal is of considerableimportance.

W. Tang et al. / Water Research 121 (2017) 302e310 303

Capacitive deionization (CDI), as a cost effective, energy efficientand environmentally friendly electrochemical technology, isattracting increasing attention for the facile removal of ionic speciesfromwater (Cohen et al., 2015; Gao et al., 2015a; Garcia-Quismondoet al., 2016; Kang et al., 2016; Suss et al., 2015; Tang et al., 2016). CDIenables ion removal at low pressures and low voltages, and has thepotential to be powered by solar energy in remote areas with theadditional possibility of energy recovery via the capacitive effectcreated on ion sorption within electrode double layers. Also, CDIoperates without the use of any added chemicals or the generationof hazardous substances, with the added possibility of recovery ofcertain ions from the concentrate stream via precipitation of highvalue minerals. Generally, a CDI cell consists of two graphite currentcollectors facilitating electron transfer, two porous carbon elec-trodes for ion sorption and a spacer channel enabling water to betransported from the perimeter to the center of the unit (Poradaet al., 2013). One relatively recent significant improvement in CDIis the inclusion of ion exchange membranes (IEMs) in front of theelectrodes in a process known as membrane capacitive deionization(MCDI). More specifically, a cation exchange membrane is placed infront of the cathodewhile an anion exchangemembrane is placed infront of the anode, allowing for counterion transport from thespacer channel into the electrode while preventing co-ions fromleaving the electrode region. In this approach, the macroporeswithin the electrodes serve as extra storage space for ions, therebyimproving the ion removal performance (Biesheuvel et al., 2011;Choi, 2014; Porada et al., 2013). To date, most (M)CDI laboratory-scale studies have focused on applying a constant charging voltagebetween the electrodes during adsorption, followed by short-circuiting or reversing the voltage to regenerate the electrodesduring desorption (Bian et al., 2015; Gao et al., 2015b; Mossad andZou, 2012, 2013). However, in practical applications, constantvoltage charging may be not appropriate considering that theeffluent ion concentrations vary greatlywith time. To obtain purifiedwater with relatively constant ion concentrations below a specifiedthreshold value, applying a constant current across the electrodesduring charging is strongly preferred.

To compare and optimize the efficacy of (M)CDI systems, variousperformance indicators have been employed (Huyskens et al., 2013;Mossad and Zou, 2012; Suss et al., 2015; Zhao et al., 2013), includingion removal efficiency, effluent ion concentrations, water recovery,charge efficiency, average salt adsorption rate (ASAR) and energyconsumption. The continuing rapid growth of CDI research neces-sitates a standardization of key performance indicators for partic-ular applications as some indicators may be inappropriate. Forexample, ion removal efficiency is mainly used in batch-mode CDItests describing the relative decrease in salt concentration in therecycled solution (Suss et al., 2015), making it unsuitable for otherCDI operational modes. With the values of energy consumptionknown, charge efficiency appears unimportant as charge efficiencyand energy consumption are closely related.

In this study, we investigate the feasibility of sulfate removal bysingle-pass constant-current MCDI in the presence of chloride as istypical in practical scenarios. Insights into understanding of theeffects of particular operational parameters (in particular, adsorp-tion/desorption current, pump flow rate and ending cell voltage) onsystem performance are provided. On the premise of maintainingthe effluent sulfate concentration below 300 mg L�1, two key per-formance indicators (i.e., water recovery and energy consumption)are used to evaluate the performance with regard to sulfateremoval. The correlation between operational parameters andperformance indicators is developed as a guide for optimization ofMCDI operation. The preferential adsorption between sulfate andchloride as affected by the selected operational parameters is alsoexamined.

2. Experimental

2.1. MCDI module

The MCDI module used in this study (Model No. VS1) wasdirectly purchased from Voltea B.V. (the Netherlands) and containsa stack of 50 cells connected in parallel electrically and with regardto fluid flow. Each cell consists of two graphite sheets as currentcollectors which are alternatingly positively and negatively biased,two porous carbon electrodes with anion or cation exchangemembranes attached to the electrode surfaces and a 200 mm thicknon-conductive nylon cloth as the spacer channel. The carbonelectrode is 200 cm2 in area and 150 mm in thickness. As the MCDImodule is commercially acquired, only limited informationregarding the exact design and manufacturing detail of the elec-trode/membrane system is available. The water is pumped into theCDI module through an opening located in one of the four cornerswith treated water flowing out from an opening in the center of theelectrode unit.

2.2. Experimental methods

The schematic diagram of the experimental setup is shown inFig. S1. The system consists of a feed vessel, a peristaltic pump(Masterflex, Cole-Parmer, USA), a MCDI module and a DC powersupply (WEP, Yihua Electronic Equipment Co., Ltd, China). In thisstudy, although convenient, dead volume was not used for mea-surement of pH and conductivity due to the great difference be-tween the measured effluent ion concentration with dead volumeand that without dead volume (further details are provided inSection S2). Prior to the test, the electrical charge remaining on thecarbon electrodes should be zero. Analytical grade sodium sulfate(Na2SO4) and sodium chloride (NaCl) were used for the preparationof feed water. All reagent solutions were prepared using 18 MU cmresistivity Milli-Q water unless stated otherwise. Initially, the MCDImodule was fully flushed using the feed water until the effluentconductivity was close to the influent conductivity. Duringadsorption, a constant electrical current was applied to the module.Once the cell voltage reached a specified value, the cell wasswitched from adsorption to desorption mode. During desorption,we employed a constant reverse electrical current (reverse-currentdesorption) until the cell voltage dropped back to 0 V. The reasonwhy reverse-current desorption instead of zero-volt desorptionwas used here is discussed in detail in Section S2. The cell voltagewas measured using an on-line voltage probe (VP-BTA, VernierCorp., USA) while the voltage readings of the power supply repre-sent the charging voltage. The data were not collected until theMCDI module had been operated for a number of cycles. Watersamples were taken from the outlet of the MCDI module atappropriate intervals and the effluent ion concentrations deter-mined using an ICS-3000 ion chromatograph (Dionex, USA).Duplicate runs were carried out for each set of experimental con-ditions and the average value presented.

In these experiments, the water recovery was calculated asfollows,

Water recovery ¼ Vp

VT¼ vads,tads

vads,tads þ vdes,tdes¼ tads

tcycle(1)

where Vp and VT (m3) are the purified water volume and total watervolume produced in one cycle, respectively. vads and vdes (m3 s�1)are the pump flow rate during adsorption and desorption, respec-tively. tads (s) is the adsorption time, tdes (s) is the desorption timeand tcycle (s) is the sum of tads and tdes. In our study, as vads was equalto vdes, a simplified expression for water recovery can be expressed

W. Tang et al. / Water Research 121 (2017) 302e310304

as tads/tcycle.The electrical energy consumption W (kJ m�3 purified water

produced) was calculated according to,

W ¼

0B@Iads,

Ztads

t0

Vch,dt þ Ides,Ztcycle

tads

Vch,dt

1CA� 10�3

.Vp (2)

where Iads and Ides (A) represent the adsorption current and thedesorption current, respectively. Vch (V) represents the chargingvoltage.

The specific ion j adsorption per cycle, Gj (mol), is derived byintegrating the difference between the feed concentration cj,feed(mM) and the effluent concentration cj,eff (mM) over time (t0/tads)and multiplying by the pump flow rate vads. The mathematicalexpression for the specific ion adsorption per cycle can be writtenas

Gj ¼ vads

Ztads

t0

�cj;feed � cj;eff

�dt (3)

3. Results and discussion

3.1. Differential separation between sulfate and chloride

To explore the differential separation between sulfate andchloride in the MCDI cells, 15 mM Na2SO4 and 15 mM NaCl wereused as the feed water under the operating conditions of adsorp-tion current 2 A, desorption current �3 A (where “minus” repre-sents polarity reversal), pump flow rate 50 mL min�1 and endingcell voltage 1.2 V. Fig. 1 shows the profiles of effluent SO4

2� and Cl�

concentrations within a full cycle. As can be seen, SO42� and Cl�

display similar adsorption and desorption behaviors, i.e., after thecommencement of the adsorption, the effluent SO4

2� and Cl� con-centrations decrease quickly and then level off, followed by a rapidincrease and attainment of an almost constant concentration dur-ing desorption. Moreover, it is found that the effluent SO4

2�

Fig. 1. Differential separation between sulfate and chloride in MCDI cells. Experi-mental conditions: sulfate feed concentration 15 mM, chloride feed concentration15 mM, adsorption current 2 A, desorption current �3 A, pump flow rate 50 mL min�1

and ending cell voltage 1.2 V.

concentration remains consistently below and above the effluentCl� concentration during adsorption and desorption, respectively,which suggests that SO4

2� is preferentially removed from theaqueous solution passing through the spacer channels in the MCDIcells. The preferential adsorption of SO4

2� over Cl� can be explainedfrom the Nernst-Planck equation describing the flux of ionsapproaching the electrode (Tang et al., 2015),Jj ¼ �Dj,ðdcj=dX þ zj,cj,df=dXÞ, where Jj is the flux of ion j (molm�2 s�1), Dj is the diffusion coefficient of ion j (m2 s�1)(DSO4 ¼ 1.07 � 10�9, DCl ¼ 2.03 � 10�9 m2 s�1) (Flury and Gimmi,2002), cj is the concentration of ion j (mM) in the spacer channeland zj is the ion charge number. Under the assumptions of gradient-less concentration profile and a linearized potential profile, the fluxratio of SO4

2� to Cl� is simplified to DSO4,zSO4

,cSO4=ðDCl,zCl,cClÞ,

which can be further simplified to DSO4,zSO4

=ðDCl,zClÞ under theconditions of equal initial concentrations of SO4

2� and Cl� in thespacer channel. The value of DSO4

,zSO4=ðDCl,zClÞ is calculated to be

greater than 1, suggesting that, compared to Cl�, more SO42� is

subject to transport from the spacer channel to the anode in acertain period of time with higher adsorption capacity in terms ofSO4

2�. While the dual anion cases may be at the simpler end ofnatural water conditions, the mathematical approach shown herecould potentially be used to qualitatively assess the preferential ionadsorption in (M)CDI for real waters (e.g., brackish groundwater)where multiple anions or cations are present at varyingconcentrations.

3.2. Characterization of electric circuit of the MCDI system

The variation of cell voltages during adsorptionwas investigatedat different adsorption currents with the results obtained shown inFig. 2a. From these results it can be observed that the cell voltageincreased rapidly at the initial stage after the adsorption currentwas applied and then gradually increased linearly. Meanwhile, anincrease in adsorption current led to a linear increase in the initialcell voltage along with a decrease in the time required to reach thefinal cell voltage. The charging voltages during adsorption atdifferent adsorption currents were also recorded and it was foundthat the difference between the charging voltage and the cellvoltage increased in direct proportion to the increase of theadsorption current (Fig. S4), implying that the external electricalresistance (Rext) in the MCDI system was constant.

A simple model was proposed to characterize the electric circuitof the MCDI system during adsorption and desorption as displayedin Fig. 2c and d. By ignoring the electrical resistance and contactinterfacial resistance of current collectors and carbon electrodes(Qu et al., 2015), the MCDI cell can be simply considered to consistof two electric double layer (EDL) capacitors and resistances of theion exchange membranes and the spacer electrolyte solution.Before supplying adsorption current to the MCDI cells, the carbonelectrodes were not charged. At the moment of connecting thecircuit, the voltage across the two EDL capacitors is zero and thevoltage across the ion exchange membranes and the spacer elec-trolyte solution represents the cell voltage. As a result, the initialelectrical resistance of the cells was calculated to be 0.099 U ac-cording to the relationship between initial cell voltage andadsorption current as indicated by the inset in Fig. 2a. The re-sistances of the ion exchange membranes and the spacer electro-lyte solution are inversely proportional to the ion concentrations inthe spacer (Biesheuvel et al., 2011; Dykstra et al., 2016), which canbe simply expressed as d/k (where d is the spacer thickness and k isthe electrical conductivity of the spacer solution) (Crow, 1994).With this formula, the cell voltage (Ucell) during adsorption isfurther given by

Fig. 2. (a) The measured cell voltages during adsorption at different adsorption currents. The inset indicates the initial cell voltage versus the adsorption current. Experimentalconditions: sulfate feed concentration 1000 mg L�1, chloride feed concentration 500 mg L�1 and pump flow rate 50 mL min�1; (b) Dynamic variation of cell voltage and chargingvoltage during adsorption and desorption. Experimental conditions: sulfate feed concentration 1000 mg L�1, chloride feed concentration 500 mg L�1, adsorption current 2 A, pumpflow rate 50 mL min�1, ending cell voltage 1.5 V and desorption current �3 A; (c, d) Equivalent circuit of the MCDI system during adsorption and desorption. Rext, Rsol, RAEM, RCEMrepresent the external electrical resistance, resistance of the solution in the spacer channel, resistance of the anion exchange membrane and resistance of the cation exchangemembrane, respectively. Canode and Ccathode represent the electric double layer capacitance of the anode and cathode, respectively.

W. Tang et al. / Water Research 121 (2017) 302e310 305

Ucell ¼ Iads,d=kþ 2Iads,t=C (4)

where C denotes the EDL capacitance of the anode or cathode.Derivatizing Eq. (4) yields

dUcelldt

¼ �Iads,dk2

,dkdt

þ 2,IadsC

(5)

where dk/dt represents the rate of change in electrical conductivityof the spacer solution over time. After an adsorption current issupplied to the MCDI cells, ions are adsorbed and the value of kdecreases and dk/dt becomes negative with its absolute valuedecreasing until it approaches zero when the ion concentrations inthe spacer become stable. Accordingly, dUcell/dt decreases and thenremains constant when dk/dt is equal to zero, which, up to thispoint, provides a satisfactory interpretation of the dynamic varia-tion in cell voltage during adsorption as shown in Fig. 2a. As for theexternal electrical resistance (Rext) of the MCDI system, it wasdetermined to be 0.1 U based on the profile of the difference be-tween the charging voltage and the cell voltage versus theadsorption current during adsorption (see Fig. S4).

In addition to enabling us to elucidate the dynamic variation ofcell voltage and charging voltage at different adsorption currentsduring adsorption, the characterization of the electric circuit of theMCDI system can also be used to explain the substantial differencebetween the cell voltage and the charging voltage during desorp-tion (Fig. 2b). Assuming that, at the moment of switching fromadsorption to desorption, the resistances of the ion exchangemembranes and the spacer electrolyte solution under the experi-mental conditions of Fig. 2b were R (U), and the voltage across thetwo EDL capacitors was U (V), then, the following equations aresatisfied according to Kirchhoff's law:

At the end of adsorption : U þ IadsðRþ RextÞ � Uch ¼ 0(6)

At the start of desorption : � U þ IdesRþ Ucell;de;0 ¼ 0

(7)

where Iads and Ides represent the adsorption current (2 A) anddesorption current (3 A), respectively. Uch represents the chargingvoltage at the end of adsorption (1.7 V). Ucell,de,0 (V) represents the

W. Tang et al. / Water Research 121 (2017) 302e310306

cell voltage at the start of desorption (0.625 V) which can be ob-tained from Fig. 2b. Substituting Iads, Ides, Rext, Uch and Ucell,de,0 intoEqs. (6) and (7), we obtain U ¼ 1.15 V, R ¼ 0.175 U. The value of R isquite reasonable considering that the initial resistances of the ionexchange membranes and the spacer electrolyte solution were0.099 U and noting that R should be larger than 0.099 U inachieving the steady-state ion concentrations in the spacer. Onreversing the DC power polarity, the two EDL capacitors would beexpected to act as a power source with charge neutralization of thecapacitors leading to a gradual decrease in the cell voltage. Asshown in Fig. 2b, this predicted behavior is consistent with theresults obtained. In addition, the voltage across the DC powersupply at the beginning of desorption is calculated tobe �Ucell,de,0 þ IdesRext ¼ �0.325 V, implying that the DC powersupply becomes an electric load component instead of a constantcurrent power source. Indeed, this is the reason why the chargingvoltage drops to 0 V until the voltage across the DC power supplyreaches a positive value.

It should be noted that the ion adsorption capacity of the carbonelectrodes depends on the voltage across the two EDL capacitorsrather than the cell voltage, whereas the energy consumption isclosely related to the charging voltage. In providing a gooddescription of the obtained voltage data, the simple circuit modelproposed here also suggests that an increase in the capacitance ofcarbon material is beneficial to enhancing the ion adsorption ca-pacity by consuming the same amount of electrical energy while adecrease in the external electrical resistance and/or the resistancesof the ion exchange membranes and the spacer electrolyte solutionis conducive to reducing the energy consumption by adsorbing thesame number of ions.

3.3. Optimization of operating parameters

For the single-pass constant-current MCDI tests under reverse-current desorption, generally, the desalination performance de-pends strongly on four operating parameters when the influent ionconcentrations are fixed, namely, adsorption current, water flowrate, ending cell voltage and desorption current. In this section,optimization of sulfate removal in the presence of chloride usingMCDI is addressed by examining the effects of the four operatingparameters on the two key performance indicators, i.e., water re-covery and energy consumption, on the premise that the effluentsulfate concentration is maintained below the specified thresholdof 300 mg L�1.

It has been reported that the effluent ion concentration could beadjusted to a certain set point only by either varying the adsorptioncurrent and/or the water flow rate for particular (M)CDI configu-rations and source waters of particular composition (Zhao et al.,2013). In preliminary studies, we found that, for a given feed so-lution of 1000 mg L�1 sulfate and 500 mg L�1 chloride, 2 Aadsorption current and 50 mL min�1 pump flow rate could reducethe effluent sulfate concentration to about 300 mg L�1. Figs. 3 and 4show the effects of adsorption current and pump flow rate,respectively, on the effluent sulfate and chloride concentrationswith the results indicating that a larger adsorption current or alower pump flow rate contributes to decreasing effluent ion con-centrations during adsorption. Despite the fact that desirable lowereffluent sulfate and chloride concentrations were achieved underthese circumstances, lower water recovery and higher energyconsumption were observed (Table 1), which is unfavorable withregard to the practical application of MCDI technology. As such,assuming that the effluent sulfate concentration is maintainedbelow the threshold of 300 mg L�1, it is more advantageous toadopt a smaller adsorption current and a higher pump flow rateresulting in higher water recovery and lower energy consumption.

That is to say, the favorable adsorption current and pump flow ratewere 2 A and 50 mL min�1, respectively. As for the effects of endingcell voltage, it was found that the steady-state effluent sulfate andchloride concentrations and the water recovery remained almostunchanged while the energy consumption decreased withdecreasing ending cell voltage (Fig. 5 and Table 1), suggesting that asmaller ending cell voltage was preferred. However, it is worthnoting that the ending cell voltage required a certain minimumvalue to ensure that the target of 300 mg L�1 effluent sulfate con-centration could be reached. In other words, switching to higherfrequency changes between the adsorption and desorption stepsinstead of charging the MCDI cells for a long period is most likely abetter operational scheme. From Figs. 3e5, it is also interesting toobserve that the adsorption current had no influence while thepump flow rate and ending cell voltage had a slight influence on thevalue of the relatively steady effluent sulfate and chloride con-centrations during desorption when the desorption current waskept constant for source waters of particular composition. Bycontrast, a more negative desorption current contributed to highereffluent sulfate and chloride concentrations during desorption asdisplayed in Fig. 6. The effluent sulfate concentration duringdesorption at a desorption current of �5 A was triple that of thefeed sulfate concentration, suggesting the possibility of recovery ofsulfate from the concentrate stream by precipitation of valuableminerals such as gypsum. In addition, with an increase in the in-tensity of the desorption current, higher water recovery and energyconsumption for the range of desorption currents examined wereobserved (Table 1), which indicates that amore negative desorptioncurrent is desirable if the electrical energy cost is not critical. Toconclude, for the case of 2 A adsorption current and 50 mL min�1

pump flow rate, on the premise that the effluent sulfate concen-tration is maintained below 300 mg L�1, the other two optimaloperating parameters were an appropriately minimized ending cellvoltage and a desorption current representing a trade-off betweenwater recovery and energy consumption.

The above discussion is based on the combination of 2 Aadsorption current and 50 mL min�1 pump flow rate. In fact, othercombinations of adsorption current and pump flow rate could alsoreduce the effluent sulfate concentration to 300 mg L�1 for a givenfeed solution of 1000 mg L�1 sulfate and 500 mg L�1 chloride. Forexample, as shown in Figs. S5�S8, under 1 A adsorption current and25 mL min�1 pump flow rate, or 4 A adsorption current and100 mL min�1 pump flow rate, 300 mg L�1 steady-state effluentsulfate concentration was reached during adsorption and the pro-files of dynamic effluent sulfate and chloride concentrations in thewhole cycle at different ending cell voltages and desorption currentbehaved similarly to those under 2 A adsorption current and50 mL min�1 pump flow rate. Here, we introduce another indicatorto characterize the differential adsorption between sulfate andchloride, i.e., sorption ratio of sulfate to chloride, which refers to theratio of sulfate adsorption to chloride adsorption per cycle. Fig. 7shows the optimization mapping of water recovery-energy con-sumption and sorption ratio of sulfate to chloride-energy con-sumption as a function of ending cell voltage and desorptioncurrent under the three examined combinations of adsorptioncurrent and pump flow rate with experimental data closer to thetop left corner considered to bemore preferable. It can be seen fromFig. 7a that, among these three combinations, the combination of1 A adsorption current and 25mLmin�1 pump flow rate resulted inthe highest water recovery and lowest energy consumption,implying that a smaller adsorption current and a lower matchingpump flow rate could be more favorable in practical applications.Similar trends in water recovery and energy consumption withincreasing ending cell voltage or desorption current were observedfor all the three combinations, i.e., increasing the ending cell

Fig. 3. Effects of adsorption current on the effluent (a) sulfate and (b) chloride concentrations on the premise of maintaining the effluent sulfate concentration below 300 mg L�1.Experimental conditions: sulfate feed concentration 1000 mg L�1, chloride feed concentration 500 mg L�1, pump flow rate 50 mL min�1, ending cell voltage 1.5 V and desorptioncurrent �3 A.

Fig. 4. Effects of pump flow rate on the effluent (a) sulfate and (b) chloride concentrations on the premise of maintaining the effluent sulfate concentration below 300 mg L�1.Experimental conditions: sulfate feed concentration 1000 mg L�1, chloride feed concentration 500 mg L�1, adsorption current 2 A, ending cell voltage 1.5 V and desorptioncurrent �3 A.

W. Tang et al. / Water Research 121 (2017) 302e310 307

voltage led to stable water recovery and higher energy consump-tionwhile increasing the intensity of desorption current led to bothhigher water recovery and energy consumption. Likewise, inFig. 7b, the combination of a smaller adsorption current and thelower matching pump flow rate contributed favorably to largersorption ratio of sulfate to chloride (i.e., larger preferentialadsorption of sulfate over chloride as the sulfate and chloride feedconcentrations were both fixed) and lower energy consumption.Meanwhile, it can be known from Fig. 7b that an increase in theending cell voltage led to a slightly smaller preferential adsorptionof sulfate over chloridewhile the desorption current had no impact.It should be noted that, here, the reason why the sorption ratio ofsulfate to chloride was only slightly above 1 is due to the lower feedmolar concentration of sulfate (10.4 mM) than chloride (14.1 mM),rather than the weak preferential adsorption of sulfate over chlo-ride in MCDI cells. Fig. 8 shows the linear positive correlation be-tween the sorption ratio of sulfate to chloride and the feed molarconcentration ratio of sulfate to chloride. If no preferentialadsorption exists between sulfate and chloride, the sorption ratio ofsulfate to chloride would increase in direct proportion to the feedmolar concentration ratio of sulfate to chloride according to y ¼ x.The experimental results indicate a slope for the sorption ratio ofsulfate to chloride as a function of feed molarity ratio of 1.58

indicating that preferential adsorption of sulfate over chlorideoccurred. In other words, 36.7% (0.58/1.58) of the adsorbed sulfatemay be attributed to the preferential adsorption of sulfate overchloride.

As suggested above, the preferential adsorption of sulfate overchloride by MCDI may be enhanced to some degree by appropri-ately adjusting the operating parameters (e.g., a smaller adsorptioncurrent and a lower matching pump flow rate) when the sulfateand chloride feed concentrations are both fixed. Nevertheless, theimproved value is still considered small compared to nanofiltration(NF). NF is an effective pressure-driven membrane process withmembrane pore size, operation pressure and cut-off ability be-tween that of ultrafiltration and reverse osmosis (Zhou et al., 2015).Important features of NF membranes include low rejection ofmonovalent ions, high retention of divalent/multivalent ions andorganic molecules above molecular weight of 300 (Lu et al., 2002;Mohammad et al., 2015). It has been reported that, for mixed saltsolutions (NaCl and Na2SO4), sulfate can be almost totally retainedby NF membranes while rejection to chloride is low and, in someinstances, negative rejection to chloride might occur (Luo andWan,2013; Tanninen et al., 2006). These properties may suggest that NFis a better choice than MCDI in treatment of brackish waters con-taining excessive sulfate. However, in fact, complete or near

Table 1Water recovery and energy consumption for single-pass constant-current MCDIunder reverse-current desorption on the premise of maintaining the effluent sulfateconcentration below 300mg L�1 for a given feed solution of 1000 mg L�1 sulfate and500 mg L�1 chloride, as a function of operating parameters varied around thereference settings: adsorption current 2 A, pump flow rate 50 mL min�1, ending cellvoltage 1.5 V and desorption current �3 A.

Operating parameters Water recovery Energy consumption (kJ m�3)

Adsorption current: pump flow rate 50 mL min�1, ending cell voltage 1.5 V,desorption current �3 A

2.0 A 64% 32062.5 A 59% 44093.0 A 52% 5286Pump flow rate: adsorption current 2 A, ending cell voltage 1.5 V, desorption

current �3 A30 mL min�1 62% 569040 mL min�1 63% 418250 mL min�1 64% 3206Ending cell voltage: adsorption current 2 A, pump flow rate 50 mL min�1,

desorption current �3 A0.6 V 66% 22000.9 V 65% 25291.2 V 64% 27821.5 V 64% 3206Desorption current: adsorption current 2 A, pump flow rate 50 mL min�1,

ending cell voltage 1.5 V�2 A 53% 3014�3 A 64% 3206�4 A 70% 3576�5 A 76% 3752

Fig. 5. Effects of ending cell voltage on the effluent (a) sulfate and (b) chloride concentraconcentration 500 mg L�1, adsorption current 2 A, pump flow rate 50 mL min�1 and desor

Fig. 6. Effects of desorption current on the effluent (a) sulfate and (b) chloride concentratconcentration 500 mg L�1, adsorption current 2 A, pump flow rate 50 mL min�1 and endin

W. Tang et al. / Water Research 121 (2017) 302e310308

complete removal of sulfate is unnecessary in most instances. Afterall, sulfate is not toxic and the maximum permissible sulfate con-centration can be as high as several hundred milligrams per litre.Recognizing this, we see (M)CDI as a promising alternative methodfor both sulfate and chloride removal with the results of this studycontributing to both fundamental understanding of the (M)CDIseparation process and insight into operating conditions suited toachieving suitable treated water quality (rather than complete ionfractionation) for a particular energy input. Admittedly, when (M)CDI is applied to treat brackish waters containing high levels oftoxic ions such as arsenate, chromate and fluoride, certain im-provements should be made to increase the extent of preferentialion adsorption, e.g., coating the carbon electrode surface with ananion-exchange resin powder with high affinity toward thecontaminant ion of concern (Kim and Choi, 2012), since themaximum permissible concentrations of these ions in treated wa-ters are low.

4. Conclusions

The field of CDI has progressed enormously in the past decadeand is currently being actively explored for water treatmentincluding brackishwater desalination andwastewater remediation.In this study, the feasibility of sulfate removal from brackish waterby single-pass constant-current MCDI under reverse-currentdesorption has been investigated. Compared to chloride, sulfate

tions. Experimental conditions: sulfate feed concentration 1000 mg L�1, chloride feedption current �3 A.

ions. Experimental conditions: sulfate feed concentration 1000 mg L�1, chloride feedg cell voltage 1.5 V.

Fig. 7. Optimization maps of water recovery-energy consumption and sorption ratio ofsulfate to chloride-energy consumption. All the three combinations of adsorptioncurrent and pump flow rate (1 A, 25 mL min�1; 2 A, 50 mL min�1; 4 A, 100 mL min�1)could reduce the effluent sulfate concentration to 300 mg L�1 for a given feed solutionof 1000 mg L�1 sulfate and 500 mg L�1 chloride.

Fig. 8. Relationship between sorption ratio of sulfate to chloride and feed molarityratio of sulfate to chloride. Experimental conditions: adsorption current 2 A, desorp-tion current �3 A, pump flow rate 50 mL min�1 and ending cell voltage 1.2 V; The totalfeed molarity of sulfate and chloride is 30 mM.

W. Tang et al. / Water Research 121 (2017) 302e310 309

was preferentially removed from the bulk solution within thespacer channel of MCDI cells. We systematically varied four inputoperating parameters, namely, adsorption current, pump flow rate,ending cell voltage and desorption current, and examined theireffects on three performance indicators (i.e., water recovery, energyconsumption and sorption ratio of sulfate to chloride) on thepremise that the effluent sulfate concentration was maintainedbelow the specified threshold of 300 mg L�1. The procedure foroptimization of the four operating parameters has been presentedwith results indicating that the combination of a smaller adsorptioncurrent and a lower matching pump flow rate was more preferableas they led to higher water recovery, larger preferential adsorptionof sulfate over chloride and lower energy consumption. The fav-oured ending cell voltage was an appropriately minimized valuewhile the preferred desorption current represented a trade-offbetween water recovery and energy consumption. As the scale-upof MCDI is readily achieved through increase in the number ofelectrode pairs in MCDI cells and/or the establishment of parallelMCDI modules, the findings obtained in this study shouldcontribute to the optimization of operating conditions when scaledup MCDI is applied to treatment of real waters containing dual or

evenmultiple anions on the condition that the concentration of oneanion (such as sulfate) is excessive and the effluent concentration ofthis anion needs to bemaintained belowa specified threshold. Suchanalyses are considered indispensable to the economic and sus-tainable implementation of MCDI technology in water treatment.

Acknowledgements

We gratefully acknowledge funding support from the AustralianResearch Council and industry partners Shenzhen Pangu Environ-mental Technologies Co Ltd and Mincarb Pty Ltd through ARCLinkage Grant LP150100854. Advice provided by Beijing OriginWater (BOW) Technical Director Dr Jing Guan and BOW colleaguesregarding the need for removal of sulfate from membrane con-centrates in wastewater treatment plants in which concentraterecycling is implemented is gratefully acknowledged.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.watres.2017.05.046.

References

Bian, Y., Yang, X., Liang, P., Jiang, Y., Zhang, C., Huang, X., 2015. Enhanced desali-nation performance of membrane capacitive deionization cells by packing theflow chamber with granular activated carbon. Water Res. 85, 371e376.

Biesheuvel, P.M., Zhao, R., Porada, S., van der Wal, A., 2011. Theory of membranecapacitive deionization including the effect of the electrode pore space.J. Colloid Interface Sci. 360 (1), 239e248.

C�alinescu, O., Marin, N.M., Ionit�a, D., Pascu, L.F., Tudorache, A., Surp�ateanue, G.,Badea, I.A., Aboul-Enein, H.Y., 2016. Selective removal of sulfate ion fromdifferent drinking waters. Environ. Nanotechnol. Monit. Manag. 6, 164e168.

Chen, Y., Wen, Y., Zhou, Q., Huang, J., Vymazal, J., Kuschk, P., 2016. Sulfate removaland sulfur transformation in constructed wetlands: the roles of filling materialand plant biomass. Water Res. 102, 572e581.

Choi, J.H., 2014. Determination of the electrode potential causing Faradaic reactionsin membrane capacitive deionization. Desalination 347, 224e229.

Cohen, I., Avraham, E., Bouhadana, Y., Soffer, A., Aurbach, D., 2015. The effect of theflow-regime, reversal of polarization, and oxygen on the long term stability incapacitive de-ionization processes. Electrochim. Acta 153, 106e114.

Crow, D.R., 1994. Principles and Applications of Electrochemistry, fourth ed. BlackieAcademic & Professional, London.

Dykstra, J.E., Zhao, R., Biesheuvel, P.M., van der Wal, A., 2016. Resistance identifi-cation and rational process design in capacitive deionization. Water Res. 88,358e370.

W. Tang et al. / Water Research 121 (2017) 302e310310

Flury, M., Gimmi, T., 2002. Solute diffusion. In: Dane, J.H., Topp, G.C. (Eds.), Methodsof Soil Analysis, Part 4, Physical Methods. Soil Science Society of America,Madison, WI, pp. 1323e1351.

Gao, X., Omosebi, A., Landon, J., Liu, K., 2015a. Enhanced salt removal in an invertedcapacitive deionization cell using amine modified microporous carbon cath-odes. Environ. Sci. Technol. 49 (18), 10920e10926.

Gao, X., Omosebi, A., Landon, J., Liu, K., 2015b. Surface charge enhanced carbonelectrodes for stable and efficient capacitive deionization using invertedadsorptionedesorption behavior. Energy Environ. Sci. 8, 897e909.

Garcia-Quismondo, E., Santos, C., Soria, J., Palma, J., Anderson, M.A., 2016. Newoperational modes to increase energy efficiency in capacitive deionizationsystems. Environ. Sci. Technol. 50 (11), 6053e6060.

Guimar~aes, D., Le~ao, V.A., 2014. Batch and fixed-bed assessment of sulphate removalby the weak base ion exchange resin Amberlyst A21. J. Hazard. Mater. 280,209e215.

Huyskens, C., Helsen, J., de Haan, A.B., 2013. Capacitive deionization for watertreatment: screening of key performance parameters and comparison of per-formance for different ions. Desalination 328, 8e16.

Kang, J., Kim, T., Shin, H., Lee, J., Ha, J.I., Yoon, J., 2016. Direct energy recovery systemfor membrane capacitive deionization. Desalination 398, 144e150.

Kim, Y.J., Choi, J.H., 2012. Selective removal of nitrate ion using a novel compositecarbon electrode in capacitive deionization. Water Res. 46, 6033e6039.

Lu, X., Bian, X., Shi, L., 2002. Preparation and characterization of NF compositemembrane. J. Membr. Sci. 210 (1), 3e11.

Luo, J., Wan, Y., 2013. Effects of pH and salt on nanofiltration�a critical review.J. Membr. Sci. 438, 18e28.

Mohammad, A.W., Teow, Y.H., Ang, W.L., Chung, Y.T., Oatley-Radcliffe, D.L., Hilal, N.,2015. Nanofiltration membranes review: recent advances and future prospects.Desalination 356, 226e254.

Mossad, M., Zou, L., 2012. A study of the capacitive deionisation performance undervarious operational conditions. J. Hazard. Mater. 213�214, 491e497.

Mossad, M., Zou, L., 2013. Study of fouling and scaling in capacitive deionisation by

using dissolved organic and inorganic salts. J. Hazard. Mater. 244�245,387e393.

Porada, S., Zhao, R., van der Wal, A., Presser, V., Biesheuvel, P.M., 2013. Review onthe science and technology of water desalination by capacitive deionization.Prog. Mater. Sci. 58 (8), 1388e1442.

Qu, Y., Baumann, T.F., Santiago, J.G., Stadermann, M., 2015. Characterization of re-sistances of a capacitive deionization system. Environ. Sci. Technol. 49,9699e9706.

Runtti, H., Luukkonen, T., Niskanen, M., Tuomikoski, S., Kangas, T., Tynj€al€a, P.,Tolonen, E., Sarkkinen, M., Kemppainen, K., R€am€o, J., Lassi, U., 2016. Sulphateremoval over barium-modified blast-furnace-slag geopolymer. J. Hazard. Mater.317, 373e384.

Silva, A.M., Lima, R.M.F., Le~ao, V.A., 2012. Mine water treatment with limestone forsulfate removal. J. Hazard. Mater. 221�222, 45e55.

Suss, M.E., Porada, S., Sun, X., Biesheuvel, P.M., Yoon, J., Presser, V., 2015. Waterdesalination via capacitive deionization: what is it and what can we expectfrom it? Energy Environ. Sci. 8, 2296e2319.

Tang, W., Kovalsky, P., He, D., Waite, T.D., 2015. Fluoride and nitrate removal frombrackish groundwaters by batch-mode capacitive deionization. Water Res. 84,342e349.

Tang, W., Kovalsky, P., Cao, B., He, D., Waite, T.D., 2016. Fluoride removal frombrackish groundwaters by constant current capacitive deionization (CDI). En-viron. Sci. Technol. 50 (19), 10570e10579.

Tanninen, J., M€antt€ari, M., Nystr€om, M., 2006. Effect of salt mixture concentrationon fractionation with NF membranes. J. Membr. Sci. 283, 57e64.

Zhao, R., Satpradit, O., Rijnaarts, H.H.M., Biesheuvel, P.M., van der Wal, A., 2013.Optimization of salt adsorption rate in membrane capacitive deionization.Water Res. 47 (5), 1941e1952.

Zhou, D., Zhu, L., Fu, Y., Zhu, M., Xue, L., 2015. Development of lower cost seawaterdesalination processes using nanofiltration technologies � A review. Desali-nation 376, 109e116.